When was the last time you made a complicated purchase? For me, it was when I was looking to buy a new TV and went down a rabbit hole of OLED, LED, Mini-LEDs and countless hours of reading user reviews, then comparing prices between retailers.
What if there’s a better way?
This week, Google made its biggest play yet to convince you there is.
At Google I/O 2026 on 19 May, Sundar Pichai unveiled what is effectively an end-to-end agentic shopping stack built on every consumer surface Google already owns. Universal Cart, a cross-merchant shopping cart that follows you across Search, Gemini, YouTube and Gmail.
The Agent Payments Protocol (AP2), which lets AI agents complete purchases on your behalf within preset spending and brand limits.
The Shopping Graph underneath it, indexing more than 60 billion product listings. And to close the loop on payments, Affirm and Klarna’s buy-now-pay-later integrated directly into Google Pay.
@google Universal Cart is your new hub for shopping on Google 🛍️ #GoogleIO♬ original sound – Google
The merchant list is just as telling. Universal Commerce Protocol partners now include Amazon, Walmart, Shopify, Nike, Sephora and Target, with checkout rolling out next in Canada and Australia, followed by the UK.
Put it all together and what you get is a single company that owns discovery, conversation, video, email, the cart, the protocol the merchants plug into, the agent that does the shopping, and the wallet that pays for it. That’s a vertically integrated agentic commerce stack, well past the experimentation phase.
For all the noise around Google’s announcement, the right way to read it is as confirmation. Agentic commerce has been quietly building across the industry for more than a year, and Google has now formally caught up to where the conversation already is.
Agentic Commerce is Bigger Than Google
In APAC alone, Visa launched its Agentic Ready programme across 10 markets with more than 50 partners at the end of April. Thirteen of those partners are in Singapore, seven in Hong Kong, and three in Malaysia, with the rest spread across Australia, Japan, South Korea, Taiwan, Thailand, New Zealand and Vietnam.
Visa Agentic Ready is, at its simplest, a programme for banks to test what happens when an AI agent, not a human, is the one making the payment, with live cards and real merchants in the loop. Which means agent-initiated payments stop being a one-bank pilot and become a collective push, rapidly speeding up adoption.
Mastercard, Stripe and OpenAI have all made their own moves in this space. McKinsey now forecasts agentic commerce could generate US$3 trillion to US$5 trillion in revenue globally by 2030.
Zac Cohen, Chief Product Officer at Trulioo, perhaps put it best during our interview:
“The laggards aren’t only going to lose potential revenue opportunities, but they might be disintermediated completely. And so now is really the time to pay attention to understand and ensure that you’re ready for the shift that’s happened.”
What’s the Missing Piece of the Puzzle?
Trust.
Every wave of digital innovation has lived or died by it. Agentic commerce is no different, except this time the question itself has changed. We’re no longer asking who is behind a transaction. We’re asking what.
This necessitates an evolution from KYC/KYB to KYA (Know-Your-Agent). Trulioo, a global digital identity company, is one of the few players in the space thinking deeply about this area. My conversation here with Zac is worth a listen if you have 10 minutes to spare.
Working on the same problem from another angle is Fime, a France-based payments testing and certification firm with more than 20 years of experience in the space. Our writer Izzat Najmi Abdullah recently sat down with their CEO, Lionel Grosclaude.
That video isn’t out yet, but a few of his talking points are worth flagging here because they fit neatly alongside Zac’s. (Subscribe to our YouTube channel to get notified when it’s out.)
His framing was that a one-time check of an AI agent is like checking IDs at the door of a nightclub. It may prove the agent was acceptable at the point of entry, but it does not tell you whether it will continue behaving properly once inside.
Fime’s FACT (Framework for Agentic Commerce Trust) is designed to monitor each transaction and check whether what the agent is buying remains consistent with the consumer’s original intent, rather than relying only on one check at the start.
So Where Does Southeast Asia Sit in All This?
Here’s the part of the Google announcement worth chewing on if you’re sitting in this part of the world. Universal Commerce Protocol-powered checkout rolls out next in Canada and Australia. After that, the UK. Nowhere in ASEAN is on Google’s near-term map.
That mismatch gets sharper when you look at how ready the consumer base actually is. A NIQ survey published with Bain in late 2025 found that 39 percent of Asia-Pacific consumers already use generative AI in online shopping, with another 40 percent saying they would.
In Indonesia and Thailand specifically, that figure climbs above 50 percent. A separate Visa-commissioned study put Singapore even higher, with roughly 77 percent of residents already using generative AI tools in daily life, and around 8 in 10 saying they rely on AI assistance when shopping online.
In other words, the demand is sitting here. The agentic checkout layer is being built somewhere else.
For ASEAN’s superapps and e-wallets, the Grabs, GoTos, GCashes and TNG eWallets, of the world, that’s a strategic problem with a narrow set of answers.
They already own most of what Google is trying to build: discovery (the app’s home screen), the merchant graph (food, transport, retail, services), payment rails, and a hold on user identity that Google does not have here. What they don’t yet have is the agentic layer sitting on top.
Three options, broadly.
Ship their own agent fast and try to be the ASEAN consumer’s primary AI interface for everything inside their walled garden. Open up to the global agentic protocols, accept reduced direct ownership of the user relationship, and compete to be the most useful merchant-side agent on someone else’s stack. Or some hybrid: a closed-loop agent inside the superapp for high-frequency local services, plus interoperability for the long tail of cross-border purchases the walled garden can’t serve.
The one option that almost certainly doesn’t work is doing nothing and assuming the moat holds. Walled gardens have held up well in Southeast Asia, but agentic commerce is the first wave that goes after the wall itself rather than what sits behind it.
Agentic commerce is not coming. It’s here.
The interesting question for everyone reading this isn’t whether to pay attention. It’s how much catching up there is to do, and how quickly.
Featured image: Edited by Fintech News Singapore, based on image by Google I/O 2026: Google Keynote